Embed Size (px)
Citation preview
2011 AMS ARAM 1
Iskenderian
Update on CoSPA Storm Forecasts
Haig Iskenderian
August 2, 2011
This work was sponsored by the Federal Aviation Administration under Air Force Contract No. FA8721-05-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the United States Government.
2011 AMS ARAM 2
Iskenderian
CoSPA Precipitation Forecast
• Deterministic forecast
• Display:
– Precipitation (VIL)
– Echo Tops
• Horizontal resolution:
– 1 km (0-2 hour)
– 3 km (2-8 hr)
• CONUS coverage
• Forecast out to 8 hours
• Time steps:
– 5 min (0-2 hr)
– 15 min (2-8 hr)
• Tactical and strategic planning tool for Traffic Flow Management
2011 AMS ARAM 3
Iskenderian
CoSPA 0-8 Hour ForecastMain Components
Corridor Integrated Weather System (CIWS)
High Resolution Rapid
Refresh (HRRR)Blending Module
0 Hour 8 Hour
Extrapolation & Nowcast Model
Numerical Weather Prediction Model
Forecast Lead
2 Hour 6 Hour
Calibration &
Phase Correction
VIL Analysis & Extrapolation Fcsts
Blending Weights (X,Y)
2011 AMS ARAM 4
Iskenderian
• CIWS extrapolation/ nowcast improvements:
– Improved satellite convective initiation
– Added Echo Top decay; improved ET growth
– Improved storm extrapolation
• HRRR improvements:
– Replaced RUC with Rapid Refresh (RR) as parent model for HRRR
– Used reduced diffusion and raised pressure top in HRRR
– Used finer resolution sea surface temperature field
• Blending improvements:
– Added regionally-varying and continually-updating calibration (“running” climatology)
– Improved convective initiation
– Improved methodology for phase correction
• Application of CoSPA to probabilistic forecasts
Outline
2011 AMS ARAM 5
Iskenderian
Nowcast ImprovementsSatellite Convective Initiation
• Incorporated NASA’s SATellite Convection AnalysiS and Tracking (SATCAST*) system in FAA CIWS
• SATCAST performs:
– Cloud classification
– Cloud tracking
– Trending of IR channels
– IR channel differencing
• 8 satellite-based interest fields indicate “confidence” in CI
• Account for cloud growth, glaciation, cloud top height
*Mecikalski and Bedka, MWR (2006)
# of SATCAST CI Indicators
Visible Satellite
8
7
6
5
4
3
2011 AMS ARAM 6
Iskenderian
Random Forest Predictor Importance
Less
Important
More
Important
Nowcast ImprovementsSatellite Convective Initiation
CI Interest
Instability Radar Growth
Solar Angle SATCAST Indicators
Fuzzy Logic CI Nowcast Model
CI within 40 km of storm
CI 40 km awayfrom storm
CIWS Forecast Engine
VIL Forecast
RF Source: John Williams, NCAR
2011 AMS ARAM 7
Iskenderian
Nowcast ImprovementsConvective Initiation Forecast Example
24 June 2011
2011 AMS ARAM 8
Iskenderian
Nowcast ImprovementsSatellite Convective Initiation
2010 CIWS
1 Hr Precipitation Forecast2011 CIWS
1 Hr Precipitation Forecast
2010 CIWS
1 Hr Binary Scores
2011 CIWS1 Hr Binary Scores
HIT MISS FALSE ALARM
Valid 19 UTC 24 June 2011
Truth
VIP Level 2 Scores
2011 AMS ARAM 9
Iskenderian
Nowcast ImprovementsEcho Top Growth and Decay
1400 Z 1600 Z1500 Z
Less
Lightning
• Combine trends in lightning and ET to improve ET forecasts
9 September 2008
2011 AMS ARAM 10
Iskenderian
ET Growth and Decay PredictorInterest Images
Filtered Echo Tops
above 30kft over TimeEcho Tops Trends Filtered Lightning
Flashes over TimeLightning Trends
Combined Trends
Filtered Trends
Time Averaged Trends
GROWTH
DECAY
50kft
45kft
40kft
35kft
30kft
25kft
20kft
15kft
10kft
5kft
0kft
2011 AMS ARAM 11
Iskenderian
Nowcast Improvements Echo Top Growth and Decay
0430+2Hr 0630
2011 CIWS: +2Hr ETF2010 CIWS: +2Hr ETF ET Truth
0430 +2Hr
0kft 5kft 10kft 15kft 20kft 25kft 30kft 35kft 40kft 45kft 50kft
HIT MISS FALSE ALARM
2 hr Verification Statistics
28 June 2010NOTE: Verification Threshold 30 Kft
2011 AMS ARAM 12
Iskenderian
Extrapolation ImprovementsMultiscale Storm Advection
Radar Track Vectors
Model Winds
Wx Classification
Spatial Interpolation
EnvelopeCell
AutomatedData Quality
Editing
Motion Data Reference vecsInput vecs
Motion Extrapolation(Advection)
Synoptic
• 13 km Circular Mean Filter• 6 Minute Correlation Time
• 69x13 km Rotated Elliptical Filter
• 18 Minute Correlation Time
• 201x101 km Rotated Elliptical Filter
• 45 Minute Correlation Time
13 km 69x13 km 201x101 km
• Multiple scales tracked (cell, envelope, synoptic), rotational & translational motion applied
• CIWS and CoSPA now share the same advection algorithm
2011 AMS ARAM 13
Iskenderian
2010 CIWS Forecast 2011 CIWS Forecast Truth
2010 CIWS +120min 2011 CIWS +120min
Level 2 B
inary
Score
s
10 May 2011 2015 UTC
2hr CSI
Scores
Dayton
AreaCONUS
2010 CIWS
21.5 15.4
2011
CIWS24.4 16.0
Extrapolation ImprovementsMultiscale Storm Advection
2011 AMS ARAM 14
Iskenderian
• CIWS extrapolation/ nowcast improvements:
– Improved satellite convective initiation
– Added Echo Top decay; improved ET growth
– Improved storm extrapolation
• HRRR improvements:
– Replaced RUC with Rapid Refresh (RR) as parent model for HRRR
– Used reduced diffusion and raised pressure top in HRRR
– Used finer resolution sea surface temperature field
• Blending improvements:
– Added regionally-varying and continually-updating calibration (“running” climatology)
– Improved convective initiation
– Improved methodology for phase correction
• Application of CoSPA to probabilistic forecasts
Outline
2011 AMS ARAM 15
Iskenderian
12Z + 9h
forecast
valid 21z
17 July
2010
HRRR ImprovementsRR Parent and Diffusion
2010RUC Parent
Strong LH6th Diffusion
2011
RR ParentStrong LH
No 6th Diffusion
NSSL Verification21 UTC
Better SE coverage
for RR parent with no 6th order diffusion
2011 AMS ARAM 16
Iskenderian
25 dBZ
40-km East CSI = .20
bias=1.50
25 dBZ 40-km East
CSI = .16
bias=1.21
12Z + 9h
forecastvalid 21z
17 July 2010
2010RUC ParentStrong LH
6th Diffusion
2011RR Parent
Strong LHNo 6th Diffusion
NSSL Verification21 UTC
HRRR:
Paper 13.2
Alexander, NOAA GSD10:45AM Thu
HRRR ImprovementsRR Parent and Diffusion
2011 AMS ARAM 17
Iskenderian
• CIWS extrapolation/ nowcast improvements:
– Improved satellite convective initiation
– Added Echo Top decay; improved ET growth
– Improved storm extrapolation
• HRRR improvements:
– Replaced RUC with Rapid Refresh (RR) as parent model for HRRR
– Used reduced diffusion and raised pressure top in HRRR
– Used finer resolution sea surface temperature field
• Blending improvements:
– Added regionally-varying and continually-updating calibration (“running” climatology)
– Improved convective initiation
– Improved methodology for phase correction
• Application of CoSPA to probabilistic forecasts
Outline
2011 AMS ARAM 18
Iskenderian
Blending ImprovementsRegionally Varying Weights
Model_WT Gen1300_Fcst05h = v1800utc
Static Weight = 0.67
Dynamic Weight = 0.35-0.71
Shifted from static to dynamic weights
• Accounts for regional variations in relative performance
• Adapts to changes in the skill of the inputs on the fly
• Improved skill of forecast
Bias, T=133, Eastern USFA2010SP2011
FA2010SP2011
Blending Source: James Pinto, NCAR
2011 AMS ARAM 19
Iskenderian
Blending ImprovementsScale Dependent Merging
• Weights are a function of time of day, leadtime, and region (where storm initiation is indicated)
– Identify model initiation areas
– Implement storm initiation weighting functions
Extrap
Model
Dash : 15 to 00 UTC
Solid : 00 to 15 UTC
Background Weights Model Weights: Storm Init
Storm Init Weights
BackgroundWeights
2011 AMS ARAM 20
Iskenderian
Latency Adj.
Summer 2010Summer 2011
Position errors determined using extrapolation forecasts instead of observations
•Less distortion of storm shapes
•Improved skill
Blending ImprovementsPhase Correction
2011 AMS ARAM 21
Iskenderian
• CIWS extrapolation/ nowcast improvements:
– Improved satellite convective initiation
– Added Echo Top decay; improved ET growth
– Improved storm extrapolation
• HRRR improvements:
– Replaced RUC with Rapid Refresh (RR) as parent model for HRRR
– Used reduced diffusion and raised pressure top in HRRR
– Used finer resolution sea surface temperature field
• Blending improvements:
– Added regionally-varying and continually-updating calibration (“running” climatology)
– Improved convective initiation
– Improved methodology for phase correction
• Application of CoSPA to probabilistic forecasts
Outline
2011 AMS ARAM 22
Iskenderian
Probabilistic Forecast Based on Weather Avoidance Field
VIL
EchoTop
CIWS WEATHER DATA
WEATHER AVOIDANCE FIELD
WAF
60km VIL Area Coverage
16
km
Ech
o T
op
Ma
xim
um
Deviation Probability
Lookup Table
DEVIATION DATABASE
Non-Deviation Deviation
CoSPA-based WAF Forecast
Time-lagged Ensemble
Data < Threshold
Data > Threshold
Center pixel location
Probability = # Pixels above threshold
Total # pixels in cylinder
WAF Fcst 1
WAF Fcst 2
WAF Fcst 3
2011 AMS ARAM 23
Iskenderian
Probabilistic ForecastDisplay Concepts
Probability ofWAF > 30% Spatial Smoothing
Sub-Sampled
and Smoothed
Processed to Approximate
Human Drawn
Convective Weather Polygons
Probabilistic Contour Underlay
2011 AMS ARAM 24
Iskenderian
Probabilistic ForecastConvective Weather Polygons
CCFP CoSPA Polygons
6 -Hr
6 -
Hr
• May provide first guess for CCFP
• CoSPA polygons being evaluated as part of
2011 Aviation Weather Center Testbed Summer Experiment
2011 AMS ARAM 25
Iskenderian
Summary
• CoSPA received upgrades to all three of its components:
CIWS extrapolation, HRRR model, and blending module
• Upgrades included improved convective initiation, storm decay, storm motion, storm structures
• Evaluation of CoSPA is taking place this summer
• Probabilistic convective polygons based on CoSPA forecasts are being evaluated as part of the AWC 2011 Testbed Summer Experiment
Acknowledgements:
J. Pinto, M. Wolfson, S. G. Benjamin, M. Steiner, S. S. Weygandt, C. Alexander, W. J. Dupree, J. K. Williams, J. R. Mecikalski, W. F. Feltz, K. Bedka, D. Morse, X. Tao, D. A. Ahijevych, C. Reiche, T. Langlois, K. L. Haas, L. J. Bickmeier, P. M. Lamey, J. M. Pelagatti, and D. D. Moradi
